// SLIDE 07 — THE "WHAT WOULD MAKE THIS WRON

THE "WHAT WOULD MAKE THIS WRONG?" QUESTION.

·"It would be wrong if any number is inaccurate." → Add accuracy criterion.

·"It would be wrong if it sounds like an endorsement instead of an analysis." → Add tone/neutrality criterion.

·"It would be wrong if it misses the third section of the report entirely." → Add completeness criterion.

·"It would be wrong if the conclusion overstates the evidence." → Add a failure criterion flagging overstatement.

·"It would be wrong if it uses private identifiers from the dataset." → Add a privacy constraint (see Chapter 3).

NARRATION

The most productive evaluation prompt design question is: what would make this output wrong? This question is uncomfortable because it requires acknowledging failure modes before they happen. But that discomfort is the point. Failure modes you can name in advance are failure modes you can prevent with criteria.

// SLIDE 11 — BRIDGE TO CHAPTER 6

BRIDGE TO CHAPTER 6.

You have now built the core scaffold: task, context, constraints, examples, and evaluation criteria. What happens when t

NARRATION

You have now built the core scaffold: task, context, constraints, examples, and evaluation criteria. What happens when the output does not meet the criteria? Chapter 6 teaches iteration and revision loops , structured patterns for diagnosing a failure, writing a targeted revision prompt, and improving the output without losing the original specification.

// SLIDE 13 — CLOSE

ASK THE QUESTION. APPLY THE FRAMEWORK.

WHAT THIS CHAPTER LETS YO//THE CORE IDEA: CRITERIA-F//FOUR TYPES OF EVALUATION

Claude Prompt Engineering · Ch.5 · Chapter 5 — Evaluation Criteria Before Output

NARRATION

That is the framework. Claude Prompt Engineering, chapter 5: Chapter 5 , Evaluation Criteria Before Output. The patterns are now in place. Apply them.